[llvm] [AArch64][GlobalISel] Improve lowering of vector fp16 fptrunc (PR #163398)

[David Green via llvm-commits]

================
@@ -536,6 +536,12 @@ G_FPTRUNC
 
 Convert a floating point value to a narrower type.
 
+G_FPTRUNC_ODD
----------------
davemgreen wrote:

Hmm. That is a new one to me and looks like an AMD-ism. We don't really have the instructions to easily support that for any other rounding modes, and it doesn't even support odd rounding modes yet. Considering we don't have a great way to conditionally legalize intrinsics like that, a separate instruction sounds like a better approach for us. (We can always change that in the future if needed).

https://github.com/llvm/llvm-project/pull/163398